Armadillo

Language: CPP

Machine Learning

Armadillo was created to bridge the gap between high-performance C++ libraries (like BLAS and LAPACK) and user-friendly numerical software like MATLAB. It provides a natural, concise API for expressing mathematical operations while relying on optimized backends (e.g., Intel MKL, OpenBLAS, LAPACK) for performance. It has become a standard tool in research, academia, and industries requiring numerical simulations and machine learning.

Armadillo is a high-quality C++ linear algebra library that provides efficient matrix and vector operations, while offering a syntax similar to MATLAB. It is designed for both speed and ease of use, making it popular in scientific computing, machine learning, and signal processing.

Installation

linux: sudo apt install libarmadillo-dev

mac: brew install armadillo

windows: vcpkg install armadillo or build from source at http://arma.sourceforge.net/download.html

Usage

Armadillo supports dense and sparse matrices, vectors, linear algebra, statistics, eigendecompositions, and integration with LAPACK/BLAS backends. Its MATLAB-like syntax makes it accessible to researchers migrating to C++.

Matrix multiplication

#include <armadillo>
#include <iostream>

int main() {
    arma::mat A = {{1, 2}, {3, 4}};
    arma::mat B = {{5, 6}, {7, 8}};
    arma::mat C = A * B;

    C.print("Result of A*B:");
    return 0;
}

Performs matrix multiplication using Armadillo with a concise MATLAB-like syntax.

Solving linear systems

arma::vec x = arma::solve(A, b);

Solves the system of linear equations Ax = b efficiently.

Eigen decomposition

arma::vec eigval;
arma::mat eigvec;
arma::eig_sym(eigval, eigvec, A);

Computes eigenvalues and eigenvectors of a symmetric matrix.

Singular value decomposition (SVD)

arma::mat U, V;
arma::vec s;
arma::svd(U, s, V, A);

Computes the SVD decomposition of a matrix.

Using sparse matrices

arma::sp_mat S(1000, 1000);
S(0,0) = 1.5;
S(100,200) = 2.3;

Efficiently handles large sparse matrices with minimal memory overhead.

Error Handling

Runtime error: matrix singular: Occurs when trying to solve a system with a singular matrix. Ensure the matrix is invertible or use pseudo-inverse.

Linker errors with LAPACK/BLAS: Ensure you link with `-llapack -lblas` or have OpenBLAS/Intel MKL installed.

Slow performance: Install optimized BLAS/LAPACK backends instead of relying on the default implementation.

Best Practices

Use Armadillo’s high-level syntax for readability, but enable optimized BLAS/LAPACK backends for performance.

Use `arma::sp_mat` for sparse matrices to save memory in large datasets.

Call `.t()` for transpose instead of manually looping.

Avoid unnecessary copies by using Armadillo’s lazy evaluation (`.eval()` when needed).

Prefer built-in solvers over manually implementing algorithms.

Official Docs Github